Robust localization against outliers in wireless sensor networks

Abstract

In wireless sensor networks, a critical system service is the localization service that determines the locations of geographically distributed sensor nodes. The raw data used by this service are the distance measurements between neighboring nodes and the position knowledge of anchor nodes. However, these raw data may contain outliers that strongly deviate from their true values, which include both the outlier distances and the outlier anchors. These outliers can severely degrade the accuracy of the localization service. Therefore, we need a robust localization algorithm that can reject these outliers. Previous studies in this field mainly focus on enhancing multilateration with outlier rejection ability, since multilateration is a primitive operation used by localization service. But patch merging, a powerful operation for increasing the percentage of localizable nodes in sparse networks, is almost neglected. We thus propose a robust patch merging operation that can reject outliers for both multilateration and patch merging. Based on this operation, we further propose a robust network localization algorithm called RobustLoc. This algorithm makes two major contributions. (1) RobustLoc can achieve a high percentage of localizable nodes in both dense and sparse networks. In contrast, previous methods based on robust multilateration almost always fail in sparse networks with average degrees between 5 and 7. Our experiments show that RobustLoc can localize about 90% of nodes in a sparse network with 5.5 degrees. (2) As far as we know, RobustLoc is the first to uncover the differences between outlier distances and outlier anchors. Our simulations show that RobustLoc can reject colluding outlier anchors reliably in both convex and concave networks.